European Journal of Remote Sensing

ISSN: (Print) 2279-7254 (Online) Journal homepage: https://www.tandfonline.com/loi/tejr20

Estimating the relationship between touristic

activities and night light emissions

Eleni Krikigianni, Chrysovalantis Tsiakos & Christos Chalkias

To cite this article: Eleni Krikigianni, Chrysovalantis Tsiakos & Christos Chalkias (2019) Estimating
the relationship between touristic activities and night light emissions, European Journal of Remote
Sensing, 52:sup1, 233-246, DOI: 10.1080/22797254.2019.1582305

To link to this article: https://doi.org/10.1080/22797254.2019.1582305

© 2019 The Author(s). Published by Informa

UK Limited, trading as Taylor & Francis
Group.

Published online: 26 Feb 2019.

Submit your article to this journal

Article views: 327

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=tejr20
EUROPEAN JOURNAL OF REMOTE SENSING
2019, VOL. 52, NO. S1, 233–246
https://doi.org/10.1080/22797254.2019.1582305

Estimating the relationship between touristic activities and night light

emissions
Eleni Krikigianni, Chrysovalantis Tsiakos and Christos Chalkias
Department of Geography, Harokopio University, Athens, Greece

ABSTRACT ARTICLE HISTORY

The spatio-temporal dynamics of urban development can effectively be considered as a Received 7 March 2018
Revised 10 February 2019
stepping stone for urban planning, decision-making and resource use and conservation.
Accepted 11 February 2019
Consecutive satellite observations of night anthropogenic lighting and their profound study
have provided beneficial estimators of both demographic and socio-economic dynamics. In KEYWORDS
the light of the above, the main aim of this paper is to examine the seasonal changes in VIIRS; night lights; OLS;
night-time satellite images, and their correlation with the touristic activity in European Union GWR; earth observations;
(EU) countries. This study is conducted using 2012 and 2013 earth observation (night light seasonal changes; human
imaginary from Defence Meteorological Satellite Program (DMSP; yearly basis) and visible activities; Europe
infrared imaging radiometer suite (VIIRS; seasonal basis) satellite programmes) and statistical
data associated with the touristic activity (total nights spent) on a country level (NUTS 0).
These data were processed using both remote sensing techniques, geographic information
system (GIS) and statistical analyses (linear regression and geographically weighted regres-
sion (GWR)). The research results show that the night-time light emissions are highly corre-
lated with the touristic activity and that the GWR proved to be an effective tool for the
investigation of this relationship. However, a number of additional parameters should be
further considered before determining the ability and the accuracy of the nightlight imagin-
ary in the assessment of the touristic activity.

Introduction been published, highlighting the applicability of Defence

Meteorological Satellite Program (DMSP)/Operational
The depiction of human settlements and activities on a
Linescan System (OLS) in monitoring spatial variations
global scale, and the estimation of its dynamics, is a
and temporal changes in an annual basis since 1992
challenging process. Although the characteristics of
(Chalkias, Petrakis, Psiloglou, & Lianou, 2005; Cinzano,
urban areas can be derived from high-spatial resolution
Falchi, & Elvidge, 2001; Doll 2000, 2008; Elvidge et al.,
satellite images, the creation of global maps on an annual
2007, 2010, 2014, 2015, 2017; Ghosh et al., 2010; Huang
basis – and even more on a seasonal basis, is not feasible
2014; Imhoff 1997; Letu, Hara, Tana, & Nishio, 2010; Lu
taking into account both data collection and analysis.
2008; Rayner, Raupach, Paget, Peylin, & Koffi, 2010;
The acquisition of socio-economic information in
Small 2005; Sutton, 1997; Sutton, 2001). However, its
various spatial scales and in an accurate and standar-
coarse spatial (2.7 km) and radiometric resolutions (6-
dized way is a challenge for in-depth policy analysis
bit quantization) in combination with the lack of on-
and decision-making. National and local authorities
board calibration, the over-glow around urban areas,
as well as the relevant European Union (EU) instru-
the saturation in the urban core and the poor geolocation,
ments (Directorates-General (DGs), European
puts barriers in its applicability in certain research areas
Economic Area (EEA), etc) require constant monitor-
and/or in the accuracy of the results (Elvidge et al., 2017).
ing of the socio-economic development across EU,
The Suomi National Polar-orbiting Partnership (S-
analysis of the influence factors and forecasting of the
NPP) satellite’s visible infrared imaging radiometer
future trends. To this end, the developments in infor-
suite (VIIRS) day/night band (DNB) sensor retains
mation technology and particularly in the domain of
important capabilities in the context of detecting and
geographic information and earth observation sys-
characterizing anthropogenic light sources. The avail-
tems provide a valuable asset towards the acquisition
ability of monthly radiometrically calibrated VIIRS
of comparable data for decision-making across EU.
DNB composites with higher spatial resolution (750
Night-time images provide a continuous, relatively
m) and digitization range (14 bit) over DMSP/OLS
accurate, affordable and direct way to identify human
generates better results regarding the detection of
activities and their spatial characteristics (Qingling,
temporal changes and trends in various phenomena
Levin, Chalkias, & Letu, 2015). Various studies have
(Miller et al., 2013).

CONTACT Eleni Krikigianni, krikigia@hotmail.com, Department of Geography, Harokopio University, Athens, Greece
© 2019 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group.
This is an Open Access article distributed under the terms of the Creative Commons Attribution-NonCommercial License (http://creativecommons.org/licenses/by-nc/4.0/),
which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
234 E. KRIKIGIANNI ET AL.
One of the most important economic activities in
EU that also falls under the scope of this paper is
tourism. Tourism has a wide-ranging impact on eco-
nomic growth, employment and social development,
and it is considered as one of the key countermea-
sures in the fight against economic decline and
unemployment (UNWTO, 2017). Any appraisal of
its competitiveness requires a good knowledge of
the volume of tourism, its characteristics, the profile
of the tourist and tourism expenditure and the ben-
efits for the European economies. Due to its impor-
tance, EU has developed an individual policy aiming
others to maintain the Europe’s positions as a leading
tourist destination and to maximize the industry’s
contribution to growth and employment (EC, 2018).
Various communications have been adopted by the
Commission in the last decade. For instance, the
communication on “A renewed EU tourism policy:
towards a stronger partnership for European tour-
ism” (European Commission (EC), 2006; COM
(2006) 134 final) pursued to address the challenges
that will shape the future of the tourism sector in EU
to develop more sustainable and environmentally
friendly tourism practices. The latter was followed
by a set of actions in relation to the sustainable
management of destinations, the integration of sus-
tainability concerns by businesses and the awareness
of sustainability issues among tourists (“Agenda for a
sustainable and competitive European tourism” (EC,
2006; COM (2007) 621 final)). In 2010, EC adopted a
communication titled “Europe, the world’s No 1
tourist destination – a new political framework for
tourism in Europe” (EC, 2006; COM (2010) 352 final)
defining a new framework for actions to increase the
competitiveness of tourism and its capacity for sus-
tainable growth.
In addition, the Commission has developed a dedi-
cated portal called Virtual Tourism Observatory (EC,
2018a) for collecting information and conducting
analysis on performance and trends in the sector.
The portal provides data visualizations as maps,
tables or graphs regarding (i) the tourist demand in
terms of change in arrivals accommodation establish-
ments, (ii) changes in the employment of the status,
(iii) net occupancy rate of rooms by month, (iv)
distribution of nights spent at tourist accommodation
establishments, (v) arrivals and expenditure of tour-
ists from non-EU countries to the EU28 and (vi)
nights spent at tourist accommodation establishments
by region. The above-mentioned analysis is based
mainly on data provided from Eurostat.
Tourism can be considered as the activity of visi-
tors to travel to a destination outside their usual
environment, for less than a year. Eurostat divides
tourism statistics in those relating to capacity and
occupancy of collective tourist accommodation and
in those relating to tourism demand. In most EU
Member States, the former information is collected
via surveys filled in by accommodation establish-
ments, while the latter are mainly collected via tra-
veller surveys at border crossings or through
household surveys (Eurostat, 2017). It must be high-
lighted that these acquisition mechanisms involve a
great deal of cost and effort and that certain countries
lack such information. Thus, at the moment, there is
not a streamlined modelling framework to support
the calculation of tourism statistics in EU countries
and regions.
Our objective is to examine the seasonal changes
in the brightness of night-time satellite images, as
well as their correlation with socio-economic activ-
ities in EU countries. More specifically, the focus lies
in the analysis of touristic activities and the investiga-
tion of the suitability of the above-mentioned earth
observation products as a proxy variable for this
domain. Moreover, the proposed analysis will contri-
bute to the study of the seasonality of touristic activ-
ities as well as the investigation/reveal of possible
arisen seasonal patterns (e.g. differences between
summer and winder periods).
Acknowledging the need for easily applicable,
efficient and low-cost collection of socio-economic
data, international organizations such as the
“World Bank” (WB), the “European Investment
Bank” (EIB) and the Organization for Economic
Co-operation and Development (OECD) identify
the importance of the use of night-time lights
images to measure and evaluate economic develop-
ment (Mukim et al., 2013).
Data and study area
Data
The main sources of data for this study are the night
lights as recorded by the VIIRS DNB sensor and the
night lights as recorded by the DMSP-OLS sensor,
which is the predecessor of VIIRS. A comparative
analysis of their main characteristics is given in
Table 1.
The substantial improvements of VIIRS images
over the DMSP can be outlined to their higher level
of quantization, their rigorous calibration and the
existences of additional spectral bands useful for
cloud, ocean and combustion source characteriza-
tions (Elvidge, Zhizhin, Baugh, & Hsu, 2015), while
their limitations can be referred to the shorter time
series and to the fact that a substantial number of
images are affected by stray light.
In the context of this paper, DMSP night light
images were obtained through the National Oceanic
and Atmospheric Administration website. The two
selected images (for years 2012 and 2013, respec-
tively) concern the average visible, stable lights and
 EUROPEAN JOURNAL OF REMOTE SENSING 235

Table 1. Comparative analysis of the main characteristics of VIIRS and OLS sensor.
Variable DMSP-OLS SNPP-VIIRS
Differences Time series availability From 1992 to 2018 From 2012-Present
Orbit Sun-synchronous-polar-850-km Sun-synchronous-polar-827-km altitude, 98.7°
altitude, 98.8° inclination, 102 min inclination, 102 min
Swath 3000 km 3000 km
Low-light imaging bandpass Panchromatic 0.5–0.9 μm Panchromatic 0.5–0.9 μm
Night-time overpass ~19:30 ~01:30
Builder/operator US Air Force NASA-NOAA Joint Polar Satellite System (JPSS)
Ground footprint 5 km × 5 km at nadir 742 m x742 m
Spatial resolution 2.7 km 750 m
Additional spectral bands Two spectral bands (visible and 21 additional bands spanning (11 at night/21 at day)
thermal infrared (TIR); 10 μm) 0.4–13 μm
Quantization 6 bit 14 bit
Saturation Common in urban cores No saturation
Calibration None for low-light imaging band On-board solar diffuser used to calibrate daytime DNB
data. Calibration is extended also to low-light
imaging mode

cloud-free coverages. The pixel value has 6-bit radio-
metric quantification levels, with a range between 0
(no light) and 63 (Wei, Hongxing, Wei, Bailang, &
Chunliang, 2014). The stable light image composites
were produced by making a further cleaning up of the
ephemeral light sources (e.g. fires, fishing boats, etc).
Their characteristics refer to cloud-free composites,
30-arc second grids (approximately 1 km), −180° to
180° longitude and −65° to 75° latitude.
The 24 monthly images (covering the period
from January 2012 to December 2013), of version
1 of the night-time VIIRS DNB, cloud-free compo-
sites and Tile 2 (75N/180W TO 0N/60W/) covering
almost the entire Europe, were accessed through
the same source. The images are produced in 15-
arc second geographic grids. Each tile is actually a
set of images containing average radiance values
and numbers of available observations. DNB
monthly composites are available in two different
configurations: The first excludes any data affected
by stray light (denoted as “vcm”), while the second
includes these data if the radiance values have
undergone the stray-light correction procedure
(denoted as “vcmsl”). Due to the reduced quality
of the “vcmsl” version, the first configuration was
used in order to achieve better results in our sea-
sonal monitoring application.
Statistical information used for this study was
acquired by The Eurostat Dissemination Database,
which provides official statistics for the EU. We
made use of the available variable regarding the
“Nights spent at tourist accommodation establish-
ments by residents/non-residents (tin001711) – Total
nights spent,” in NUTS 0 level (country level), on a
monthly basis, for the years 2012 and 2013, respec-
tively (Eurostat, 2018). Taking also into account that
nights are used to assess the flows of visitors (OECD,
2018), we consider that “total-nights spent” value can
be used as a reliable index of “touristic activity” at a
1
The alphabetical part of the code stands for tourism industry, and it refers to the datas-et identification code, which
was used for this study (Eurostat, 2018).
national scale (Papatheodorou & Arvanitis, 2014). It
must be also taken into consideration that the statis-
tical data are also affected by tax evasion (establish-
ments which do not declare the real number of stays)
and by current accommodation trends (such as
Airbnb, couch-surfing), which do not make their
data publicly available (Stathakis & Baltas, 2017).
Study area
Europe is selected as a study area for this analysis.
According to EC (Eurostat, 2013), Europe is consid-
ered as a prominent tourist destination and world’s
most visited region, holding approximately a 49%
share of the global tourist arrivals in 2016
(UNWTO, 2017). Given also the fact that tourism,
except for its operation as an engine of economic
prosperity, is considered as a significant source of
employment –especially in the Mediterranean
region (EC, 2001) – and its rapid growth during
the last 15 years (Stevenson, Aiery, & Miller, 2008),
its study through the use of satellite products and its
consideration as a proxy variable can provide policy-
makers and tourism industry with a comparative
advantage in the way of defining new sustainable
strategies.
Applied method
In this study, a two-step methodology is applied, for
examining the seasonal changes in night images and
their correlation with the touristic activity. The first
step comprises a sequence of image-processing ana-
lyses of DMSP and VIIRS images, respectively, lead-
ing to the extraction of the sum of lights (SOL) index.
The second step is related to the statistical analysis,
using the linear and the geographically weighted
regression (GWR). The aforementioned steps are
thoroughly described in the following sections.
236 E. KRIKIGIANNI ET AL.

Image processing be to set these pixels as not a number (NaN), it is

We started by projecting the images to the Lambert
azimuthal equal-area (ETRS89-LAEA Europe; EPSG:
3035) system. DMSP images were then calibrated
using a second-order polynomial regression model
(1), and the intercalibration model coefficients (Table
2) of each image have been calculated by Galimberti
(2017):
Imagecalibrated ¼ C2  Image2 þ C1  Image þ C0
In order to avoid both the blooming effect and the
light saturation problem, the digital number (DN)
values of the calibrated images were based on the 6-
bit (0–63) quantization. Thus, zero value pixels and
pixels with a significant lower value (DN < 6) were
discarded, and blooming effect was removed, and
brightest pixels (DN = 63) were removed, and satura-
tion problem was resolved. Given this, the threshold
for the lowest value has been set to zero and for the
highest to 63, for each of the two calibrated images.
For VIIRS, a mean algorithm is used to remove the
abnormal lights (outliers) that may be associated with
fires and gas flares. The maximum DN value of the
five biggest EU28 cities (London (UK13), Berlin
(DE30), Rome (ITI4), Paris (FR10) and Madrid
(ES30)) was calculated and assigned as the threshold
value, based on which the outliers will be removed
(Tables 3 and 4). For pixels exceeding this threshold
value, the mean algorithm is used, calculating the
new value based on the surrounding pixel values in
5 × 5 block area. However, the outliers of the image
are not removed via this single application of the
algorithm while they correspond to a quite small
fraction of the whole image. While an option could
Table 2. Intercalibration model coefficients.
Image C0 C1
F182012 1.6511 0.3815
F182013 1.5803 0.4479
preferred to iterate the algorithm multiple times
(until the complete removal of the abnormal values)
due to the fact that these pixels are located in the
centre of large cities and it would be “un-natural” to
have blank spots. Finally, the negative values of the
images have been set up to 0.
Following this preliminary correction, two
approaches are basically applied in the literature, in
order to remove the background noise in DNB
(1) monthly composite data. The first one is referred as
the noise masking method (NMM) and involves the
removal of the background noise via a mask gener-
ated by DMSP/OLS stable lights that is applied to
DNB composite data (Shi et al., 2014a, 2014b; Jing,
Shao, Cao, Fu, & Yan, 2016; Li, Huimin, Xiaoling, &
Chang, 2013). The DNB data are resampled to the
same resolution as DMSP (30 arc seconds). Then, the
pixels of DMSP stable lights with a positive value
(DN > 0) are extracted in order to generate a mask.
The pixels of the DNB data that fall outside the mask
are set as NaN, while the pixel value is kept the same
for pixels inside the mask (Jing et al., 2016).
Following a similar approach, Li et al. (2013) multi-
plied the mask of DMSP/OLS data with the VIIRS
image in order to derive noise-free VIIRS data.
The basic drawback of this method is the fact that
relies on the DMSP stable lights to generate the mask.
The lack of on-board calibration in combination with
the saturation in areas of intense brightness and the
blooming effect may lead to inaccurate results. In
addition, the mask will exclude some low-light emis-
sion sources such as small towns and road features
that the DNB product is sensitive enough to pick up.
Thus, Jing et al. (2016) introduced the optimal
threshold method to remove background noise. This
method relies on the use of an object function to
determine the optimal threshold. Given the fact that
DMSP/OLS studies have already proved the close
C2 R2 relationship between the light intensity and the type
0.0078 0.954 of land use/cover, the latter attempt to use the
0.0064 0.957

Table 3. Maximum DN values for VIIRS monthly composites for 2012.

NUTS 2 areas Apr May June July Aug Sept Oct Nov Dec
UKI3 0 0 0 0 0 195.27 314.35 186.96 314.27
DE30 0 0 0 0 0 250.75 109 168.23 123.29
ES30 218.12 0 0 0 172.08 20.67 413.27 307.42 313.19
ITI4 181.39 0 0 0 703.9 186.57 195.72 187.36 286.47
FR10 0 0 0 0 0 264.44 202.49 223.72 265.72

Table 4. Maximum DN values for VIIRS monthly composites for 2013.

NUTS 2 areas Jan Feb Mar Apr May June July Aug Sept Oct Nov Dec
UKI3 92.78 258.46 245.98 161.43 0 0 0 0 121.10 143.31 313.62 202.35
DE30 39.31 96.08 160.02 0 0 0 0 0 0 145.17 74.07 126.06
ES30 176.70 240.93 193.93 212.76 454.70 0 0 181.73 196.5 351.40 217.92 334.84
ITI4 182.02 226.16 147.18 199.56 0 0 0 196.21 171.03 184.22 206.88 208.22
FR10 259.88 321.85 225.55 230.82 0 0 0 0 228.36 203.14 515.55 250.28

correlation between light area and built-up area as the
object function.
In the context of this analysis, the background
noise will be removed via the well-established corre-
lation of night light emissions with the population.
The radiance values ranging from 0 to 1 × 10¯9 W/
cm2 sr and with pace of 0.1 are used as potential
threshold values. Then, the SOL for each region is
calculated for each of the selected threshold values.
The correlation coefficient is calculated between the
population of January 2013 and the EU countries.
The threshold value with the highest R2 is used to
remove background noise. In our case, pixels with
value smaller than 1 nW/cm2 sr are removed from
the monthly composites.
As mentioned before, a number of VIIRS monthly
composites face the problem of being affected by
stray light. As a consequence, composites of April,
May, June, July, August and September contain pixels
with no data, because of the solar illumination which
affects the northern regions in summer (Zhao, 2017).
Computational techniques were applied so as to fore-
cast the no-data values in the images of the above-
mentioned months.
In the context of this analysis, linear regression
and exponential smoothing have been used and
evaluated towards their applicability in calculating
the missing values in the VIIRS imagery. The linear
regression method is used to investigate the poten-
tial relationship between a variable of interest (often
called the response variable, but there are many
other names in use) and a set of one of more vari-
ables (known as the independent variables or some
other term).
Forecasts are obtained via the use of a simple
linear model (2) as follows:
Y ¼ β0 þ β1 x (2)
where x is the value of the predictor for which we
require a forecast (i.e. pixels with known values) and
y is the value of the corresponding forecast (pixels
with missing values).
When this calculation is done using an observed
value of x from the data, we call the resulting value
of y a “fitted value.” This is not a genuine forecast
as the actual value of y for that predictor value was
used in estimating the model, and so the value of y
is affected by the true value of y. When the value of
x is a new value (i.e. not part of the data that were
used to estimate the model), the resulting value of y
is a genuine forecast (Hyndman & Athanasopoulos,
2013).
The second method implies the use exponential
smoothing. Forecasts produced using exponential
smoothing methods are weighted averages of past
observations, with the weights decaying exponentially
as the observations get older. In other words, the
EUROPEAN JOURNAL OF REMOTE SENSING 237
more recent the observation, the higher the associated
weight. This framework generates reliable forecasts
quickly and for a wide spectrum of time series
which is a great advantage and of major importance
to applications in industry. The simplest of the expo-
nentially smoothing methods is naturally called “sim-
ple exponential smoothing” (SES). (In some books, it
is called “single exponential smoothing.”) This
method is suitable for forecasting data with no
trend or seasonal pattern (Hyndman et al., 2014). In
the context of this study, the Holt–Winters exponen-
tial smoothing function was utilized.
Original monthly VIIRS DNB composites of
January, February and March were stacked sequen-
tially and then each pixel had three time series DN
values applied. The following exponential smoothing
(Equation (3)) described by Holt (2004) is used in the
context of this analysis:
St ¼ axt þ ð1  aÞSt1 ð0<a<1Þ (3)
where x denotes an observation value (i.e. the radi-
ance of a pixel in an original monthly VIIRS DNB
image), S represents a smoothed (or predicted) value,
t denotes a period (i.e. a month in this study) and α is
the smoothing factor. The above-mentioned equation
demonstrates that in exponential smoothing, a new
smoothed (or predicted) value St is affected by a
current observation value xt and the last period
smoothed value St – 1. Old observation values have
an exponentially declined effect on the predicted
value depending on period t. The value of α deter-
mines the smoothing effect. Large values of a (i.e.
close to 1) contribute to a small smoothing effect,
while smaller values of a (i.e. close to 0) indicate
that the prediction is influenced considerably by
both current and the previous observations.
Usually, a region’s economic level and population
in 1 month are closer to its adjacent months than in
later/earlier months. For instance, night-time lights in
April should be closer to the brightness levels of
March than the ones in February or January.
Since in the original April image, only a portion of
pixels suffer from no data values, a patched image for
April was produced by combining the original and the
estimated April image. The patched April image was
added into the stacked images of January, February
and March resulting in each pixel having four time
series radiance values. The same process was repeated
for the images of May and June. Following the same
approach, the original images of December, November
and October were sequentially stacked to produce a
patched image of September. The patched image of
September was added into the stacked images of
December, November and October so as to produce
a patched image of August. Similarly, the image of July
was calculated.
238 E. KRIKIGIANNI ET AL.
To support the evaluation of the Holt–Winters
forecasting method, the image of April 2013 is calcu-
lated for the Municipality of Athens, as also performed
in the linear regression. A visual comparison between
the forecasted images from exponential smoothing and
linear regression and the original image can be
achieved in Figure 1.
It seems that the linear regression method pro-
vides results closer to the original image of April
than the one of exponential smoothing. However, in
order to obtain more accurate results, the Pearson
correlation is calculated for each of the forecasted
images with the original image (Table 5). In particu-
lar, a set of 100 random points was created in which
values of cells from the three images were stored, for
the defined locations. The linear regression appears
higher values of R2 (0.9142) in comparison with the
Holt–Winter’s exponential smoothing. Thus, the lin-
ear regression method is used for the calculation of
no-data values in the monthly composites.
For the calculation of the final images, the linear
regression method has been applied. Thus, forecasted
and patched images for April, May, June, July, August
Figure 1. Visual comparison of the forecasted images from exponential smoothing and linear regression with the original image
(in DN values).
Table 5. Evaluation of the exponential smoothing and linear regression models – calculation of Pearson correlation.
Pearson Correlation
Variables M042016_Original
M042016_Original 1.00
M042016_Exponential Smoothing 0.89
M042016_Linear Regression 0.91
and September were produced for the preprocessed
monthly composites. The new produced monthly
composites were resampled to the cell size (436 ×
436) of the existing raster images. The new produced
composites were constituted both from the new fore-
casted pixel values and the existing ones – where
available – using raster mosaic techniques.
The last step of this process was the calculation of
the SOL index (both for DMSP and VIIRS images)
within the zones of European countries (NUTS 0
level). SOL was calculated per country, both on a
yearly basis and on a semester (6-month) basis for
the years 2012 and 2013, respectively.
Statistical analysis
Exploring the possibility of using night light imagery
as a proxy for monitoring touristic activities demands
the implementation of regression models that will
establish the relationship between SOL index and
the variables under investigation. Regression analyses
attempt to show the degree to which one or more
variables can potentially cause positive or negative
M042016_Exponential Smoothing M042016_Linear Regression
0.89 0.91
1.00 0.96
0.96 1.00

change in another variable (Murak, 2013). The com-
monly used linear model (4) is adopted (Darlington
& Hayes, 2017):
Y ¼ b0 þ b1X1 þ b2X2 þ ::: þ bkXk þ e (4)
where Y is the dependent variable that we are trying
to predict or understand (in our case the total nights
spent), X is the independent variable of the model
(the SOL index), b0 is the regression constant (also
called Y intercept), b are the regression coefficients or
simply the regression weight that determines how
much the equation uses values on that variable to
produce an estimate of Y and e is the random error
(residuals) that indicates the unexplained portion of
the dependent variable.
Since the derivation of the regression equation is
based on minimizing the sum of the squared resi-
duals, this method is called ordinary least squares
regression or just OLS regression. In OLS regression,
the distribution’s spread is measured via the standard
deviation, while the correlation between X and Y is
calculated via the Pearson correlation coefficient. The
Pearson correlation coefficient (5), or simply the cor-
relation, between X and Y is defined as the covariance
of X and Y divided by the product of their standard
deviations.
rxy ¼ Cov ðXY Þ=Sx Sy (5)
To further examine the spatial heterogeneity and the
variance of the residuals, a local model, namely a GWR
model, with adaptive spatial kernels has been employed.
The GWR is the first alternative approach to overcome
the lack of spatial stability (Fotheringham & Charlton,
1998). It is a variation of single or multiple linear regres-
sion. Its difference lies in the fact that the observations are
weighted by their geographic location. This is in supra-
scale as a direct result in the analysis of classical regression
while in the GWR in local scale (Fotheringham,
Brunsdon, & Charlton, 2002; Milaka, 2010). The formula
of the GWR is as follows:
Yi ¼ βoi þ jχijβjðρiÞ þ εi (6)
where ρi is the geographic location of observation i. A
fundamental idea of GWR is the calculation of the
parameters βj (ρi) for each variable j and for each
spatial unit i (summed βij) (Fotheringham et al.,
2002). The Akaike Information Criterion (AIC)
method is used to calculate the bandwidth. The
input features (dependent and explanatory variables)
are the same with those specified in the OLS models.
Results
A cross-sectional (linear) regression analysis has been
applied for defining the correlation between the SOL
of DMSP images and the variable of “total nights
spent in a place” on an annual basis (for 2012 and
EUROPEAN JOURNAL OF REMOTE SENSING 239
2013, respectively). The calculated Pearson correla-
tion (0.94 for 2012 and 0.93 for 2013) indicates the
positive correlation of the examined variables and
implies that a liner equation describes the relation-
ship between the examined variables.
The R value (coefficient of correlation) represents
the simple correlation and is 0.943 (for 2012) and
0.939 (for 2013), which indicates a very high degree
of correlation between the examined variables. The R2
(coefficient of determination) indicates the percen-
tage of variability of SOL interpreted by the model.
Possible values range from 0 to 1. Values closer to 1
indicate that the model has a better predictive char-
acter. In our case, this percentage is 89% (for 2012)
and 88% (for 2013), which is very high too. Results
from the statistical analysis using IBM SPSS, indicate
that the regression model predicts the depended vari-
able, significantly well (p = 0.0000, p < 0.0005), and
there is a positive correlation between the two vari-
ables (b1 = 130 for 2012 and b1 = 124 for 2013). A
schematic presentation of the linear regression model
using the OLS (Figures 2 and 3), indicates also the
absence of a specific structure (e.g. grouping) in over-
estimations (areas depicted in the shades of red col-
our) and in underestimations (areas depicted in the
shades of blue colour), thus the satisfactory perfor-
mance of the model is depicted.
For further examination of the spatial relationships
and exploration of potential geographical heterogene-
ities, Moran’s I index has been calculated. When the
index gets values higher than 0, then the set of obser-
vations shows a grouped spatial pattern, while for less
than 0, a scattered pattern is presented. Given the z-
score (0.974469 for 2012 and 0.951930 for 2013) and
p-value (0.3298 for 2012 and 0.3411 for 2013), it is
assumed that the residuals’ pattern does not appear to
be significantly different from random. DMSP analysis
indicates positive results regarding the ability of the
model to interpret the variables under investigation.
Continuing with VIIRS, a similar cross-sectional
(linear) regression analysis (global model) has been
applied for defining the correlation between the SOL
of VIIRS images and the variable of “total nights
spent in a place” on a semester (6-month) basis and
more specifically between April–September and
October–March (for 2012 and 2013, respectively).
The results of the statistical analysis are shown in
Table 6.
Given the z-score of 3.0761 (for April–September
2012) and 2.6629 (for April–September 2013), there is
a less than 1% likelihood that this clustered pattern
could be the result of random chance. Also, the z-
score of 1.8931 (for October–March 2012) indicates
that there is a less than 10% likelihood that this
clustered pattern could be the result of random
chance. However, for October–March 2013 although
the given z-score of 1.5274 reflects a pattern which
240 E. KRIKIGIANNI ET AL.

Figure 2. Ordinary least squares for SOL 2012 (predictor variable) and total nights spent (response variable) – (residuals) using DMSP.

Figure 3. Ordinary least squares for SOL 2013 (predictor variable) and total nights spent (response variable) – (residuals) using DMSP.
 EUROPEAN JOURNAL OF REMOTE SENSING 241

Table 6. OLS statistical results for VIIRS 2012 and 2013.
Summer Summer Winter
Indexes period 2012 period 2013 period 2012 period 2013
OLS
R2 0.82 0.72 0.65
Pearson 0.90 0.87 0.83
Moran’s I 0.129808 0.107529 0.064978
z-score 3.076078 2.662905 1.893164
p-value 0.002097 0.007747 0.058336
AICc 1030.53 1038.11 1000.44
does not appear to be significantly different from
random, the R2 (68%) indicates not quite high degree
of correlation between the examined variables.
Interpretation of results
Individual results of local R2, which indicate how well
(values close to 1.0) the local regression model fits
observed y values (values close to 0.0 indicate that
the local model is performing poorly), are presented
in Table 7.
Following the execution of GWR, higher values of
the local coefficient of determination R2 are depicted
in the southern countries of Europe for both the
summer and winter periods; this indicates that the
method performs better in Southern Europe rather
than in Northern Europe. These values are system-
atically higher for the summer period analysis indi-
cating stronger correlation during the summer
period, while relatively lower values of local R2 are
observed in north Europe (Figure 4–7).
The diagnostic statistics, which are derived from
ArcGIS and IBM SPSS, provide a good indication of
the existence of possible goodness of fit of the model.
Any spatial dependences which were presented in the
residuals in the global model have been removed with
the geographical weighting in the local model
(Moran’s I index). The local R2 takes slightly higher
values, and this is a good token of improvement in
the model performance. Also, by comparing the glo-
bal model’s AICc value with the local model’s AICc
value, it is evident that in three of the four cases, the
lower values in the local model depict a strong evi-
dence of an improvement in the fit of the model to
the data, namely a better performance for the local
model. This improvement can also be statistically
supported and verified by the calculated Moran’s I
index in the local model, which shows lower
variances and higher probabilities of random distri-
Winter bution (p-values and z-scores).
The relationship between nights spent in a place
and DMSP generated SOL is positive, suggesting areas
0.68
0.87 with higher SOL are related to touristic activities, in
0.045629 terms of nights spent in these areas. The global model
1.527449
0.126649 of VIIRS appears to follow this positive correlation
1003.37 between the examined variables. Besides, the local
regression model shows that this relationship is stron-
ger in Southern Europe during the summer, where
there is a peak in touristic activity. From the above,
it seems that the association between touristic activities
and night-time emissions, although evident, is not
homogeneous in both spatial and temporal terms.
Seasonality of touristic activity can be captured by
seasonal night light images. Further to this, results
also show that the nights spent in a place have better
correlation with SOL during the summer period
(April–September), in comparison with the winter
period (January–March and October–December),
where the correlation is lower (<80%) but still in
high level.
Conclusions
In this study, we investigate the seasonal changes in
the brightness of night-time satellite images, as well
as their correlation with the touristic activity in EU
countries. Using a linear regression model, DMSP
examination on a yearly basis showed a strong posi-
tive relationship between the variables under investi-
gation. VIIRS global examination showed that the
positive relationship still exists on a semester (6-
month) basis, but the level of correlation is different
between the winter and summer periods. The local
model (GWR) in VIIRS produced better results in
terms of goodness of fit of the model and residuals
independency than the global model (OLS). Also,
from the local regression analysis, it can be assumed
that the model works better for the South Europe.
Recommendation and future research
It is worth mentioning that in future researches, pos-
sible use of complete data series of light emission
(without forecasted images), as well as the simulta-
neous investigation of other variables (such as popula-
tion and/or gross domestic product (GDP) or energy

Table 7. GWR statistical results for VIIRS 2012 and 2013.

Indexes Summer period 2012 Summer period 2013 Winter period 2012 Winter period 2013
GWR
R2 0.32–0.92 0.31–0.95 0.30–0.88 0.29–0.86
Moran’s I 0.051305 0.029315 0.013380 −0.004890
z-score 1.610853 1.198801 0.930152 0.574600
p-value 0.107212 0.230605 0.352292 0.565562
AICc 1497.048 1023.79 997.57 1001.17
242 E. KRIKIGIANNI ET AL.

Figure 4. Mapping of local R2 for summer (April–September) period 2012, using VIIRS SOL 2012 predictor variable and total
nights spent response variable.

Figure 5. Mapping of local R2 for summer (April–September) period 2013, using VIIRS SOL 2013 predictor variable and total
nights spent response variable.
 EUROPEAN JOURNAL OF REMOTE SENSING 243

Figure 6. Mapping of local R2 for winter (January–March and October–December) period 2012, using VIIRS SOL 2012 predictor
variable and total nights spent response variable.

Figure 7. Mapping of local R2 for winter (January–March and October–December) period 2013, using VIIRS SOL 2013 predictor
variable and total nights spent response variable.
244 E. KRIKIGIANNI ET AL.
consumption), could interpret better the relationship
between the examined variables and could give better
results in the assessment of the model. Besides con-
sidering also that the examined socio-economic vari-
able (total nights spent) has a strong relationship with
the space, a further investigation of its correlation with
the SOL index is highly recommended, by taking into
account categories of areas around Europe (geographi-
cal heterogeneity), for example, North and South
Europe, large cities (inland) and coastal zones, and/or
examining the level of their correlation in other spatial
classification (e.g. NUTS 1 or NUTS 2). Thus, a stra-
tified more detailed spatially analysis could provide
more reliable results.
In addition, the existence of other parameters that
affect the recording of night light emission sources
(such as albedo, land cover, etc) (Liang et al., 2005)
should also be investigated, in order to determine and
specify the ability and the accuracy of the night light
imaginary in the assessment of the touristic activity.
Acknowledgments
Version 4 DMSP-OLS Night-time Lights Time Series. Image
and data processing by NOAA’s National Geophysical Data
Centre. DMSP data collected by US Air Force Weather Agency.
Version 1 Night-time VIIRS Day/Night Band Composites pro-
duced by the Earth Observations Group (EOG) at NOAA/
NCEI.
Disclosure statement
No potential conflict of interest was reported by the authors.
References
Chalkias, C., Petrakis, M., Psiloglou, B., & Lianou, M.
(2005). Modelling of light pollution in suburban areas
using remotely sensed imagery and GIS. Journal of
Environmental Management, 79, 57–63. doi:10.1016/j.
jenvman.2005.05.015
Cinzano, P., Falchi, F., & Elvidge, C. (2001). The first world
atlas of the artificial night sky brightness. Monthly
Notices of the Royal Astronomical Society, 328(3), 689–
707. doi:10.1046/j.1365-8711.2001.04882.x
Darlington, B.R., & Hayes, F.A. (2017). Regression analysis
and linear models: Concepts, applications, and implemen-
tation (methodology in the social sciences). The Guilford
Press. ISBN 978-1-4625-2113-5
Doll, C.H., Muller, J.P., & Elvidge, C.D. (2000). Night-time
imagery as a tool for global mapping of socioeconomic
parameters and greenhouse gas emissions. AMBIO: A
Journal of the Human Environment, 29, 157–162. 2000
doi:10.1579/0044-7447-29.3.157
Doll Christopher, N.H. (2008). CIESIN thematic guide to night-
time light remote sensing and its applications. New York, NY:
Center for International Earth Science Information Network
(CIESIN) Columbia University Palisades.
Elvidge, C., Baugh, K., Sutton, P., Bhaduri, B., Tuttle, B.,
Tilottama, G., … Erwin, E. (2010). Who’s in the dark:
Satellite based estimates of electrification rates. Earth
observation group, NOAA National Geophysical Data
Center (pp. 102–123). doi:10.1177/1753193410380585
Elvidge, C., Keith, D.M., Tuttle, B.T., & Baugh, K.E. (2010).
Spectral identification of lighting type and character.
Sensors, 10(4), 3961–3988. doi:10.3390/s100403961
Elvidge, C., Zhizhin, M., Baugh, K., & Hsu, F. (2015).
Automatic boat identification system for VIIRS low light
imaging data. Remote Sensing. doi:10.3390/rs70303020
Elvidge, C., Zhizhin, M., Baugh, K., Hsu, F., & Ghosh, T.
(2015). Methods for global survey of natural gas flaring
from visible infrared imaging radiometer suite data.
Energies, 9(1), 14. doi:10.3390/en9010014
Elvidge, C.D., Hsu, F.-C., Baugh, K.E., & Ghosh, T. (2014).
National trends in satellite-observed lighting 1992-2012. In
Q. Weng (Ed.), Global urban monitoring and assessment
through earth observation (pp. 96–120). Boca Raton: CRC
Press.
Elvidge, C.D., Baugh, K., Zhizhin, M., Hsu, F.C., & Ghosh,
T. (2017). VIIRS night-time lights. International Journal
of Remote Sensing, 38(21), 5860–5879. doi:10.1080/
01431161.2017.1342050
Elvidge, C.D., Cinzano, P., Pettit, D.R., Arvesen, J., Sutton,
P., Small, C., … Ebener, S. (2007). The Nightsat mission
concept. International Journal of Remote Sensing, 28(12),
2645–2670. doi:10.1080/01431160600981525
European Commission (EC). (2001). Tourism trends in
Mediterranean Countries Retrieved January 30, 2018,
from http://ec.europa.eu/eurostat/documents/3217494/
5628504/KS-40-01-666-FR.PDF/ff0abde3-19a9-4ea5-
b7bf-11fa860a3289?version=1.0
European Commission (EC). (2006). Communication from the
commission: a renewed EU tourism policy: Towards a stron-
ger partnership for European Tourism. Retrieved January
30, 2018, from http://eur-lex.europa.eu/legalcontent/EN/
TXT/PDF/?uri=CELEX:52006DC0134&from=EN
European Commission (EC). (2018a). Tourism. Retrieved
January 30, 2018, from https://ec.europa.eu/growth/sec
tors/tourism_en
European Commission (EC). (2018b). Virtual tourism
observatory. Retrieved January 30, 2018, from https://
ec.europa.eu/growth/tools-databases/vto/
Eurostat. (2013a). Regions in the European Union -
Nomenclature of territorial units for statistics - NUTS
2013/EU-28. Retrieved January 30, 2018, from http://ec.
europa.eu/eurostat/documents/3859598/6948381/KS-GQ-
14-006-EN-N.pdf/b9ba3339-b121-4775-9991-
d88e807628e3
Eurostat. (2013b). The EU in the world 2013 – A statistical
portrait. Retrieved January 30, 2018, from http://ec.
europa.eu/eurostat/documents/3217494/5748537/KS-30-
12-861-EN.PDF/d60278b9-d0ee-47c1-bb96-
2dbb9a5b80cb?version=1.0
Eurostat. (2017). Tourism statistics. Retrieved January 30, 2018,
from http://ec.europa.eu/eurostat/statistics-explained/index.
php/Tourism_statistics#Data_sources_and_availability
Eurostat. (2018). Data (Total nights spent). Retrieved
January 30, 2018, from http://ec.europa.eu/eurostat/
tgm/table.do?tab=table&init=1&language=en&pcode=
tin00171&plugin=1
Fotheringham, A., & Charlton, M. (1998). Geographically
weighted regression: A natural evolution of the expan-

sion method for spatial data analysis. Environment and
Planning A, 1905–1927. Retrieved January 30, 2018,
from http://epn.sagepub.com/content/30/11/1905.
abstract?id=a301905 doi:10.1068/a301905
Fotheringham, A.S., Brunsdon, C., & Charlton, M. (2002).
Geographically weighted regression: The analysis of spatially
varying relationships. Chichester: John Wiley and Sons.
Galimberti, J.K. (2017). Forecasting GDP growth from the
outer space. Retrieved January 30, 2018, from https://
www.research-collection.ethz.ch/handle/20.500.11850/
128755
Ghosh, T., Powell, R., Elvidge, C., Baugh, K., Sutton, P., &
Anderson, S. (2010). Shedding light on the global distribu-
tion of economic activity. The Open Geography Journal,
2010(3), 148–161. doi:10.2174/1874923201003010147
Holt, C.C. (2004). Forecasting seasonals and trends by
exponentially weighted moving averages. International
Journal of Forecasting, 20(1), 5–10. doi:10.1016/j.
ijforecast.2003.09.015
Huang, Q., Yang, X., Gao, B., Yang, Y., & Zhao, Y. (2014).
Application of DMSP/OLS nighttime light images: A
meta-analysis and a systematic literature review. Remote
Sensing, 6(8), 6844–6866. doi:10.3390/rs6086844
Hyndman, J.R., & Athanasopoulos, G. (2013). Forecasting:
Principles and practice. Melbourne, Australia: OTexts.
ISBN 978-0-9875071-0-5.
Hyndman, R.J., Athanasopoulos, G., Razbash, S., Schmidt,
D., Zhou, Z., Khan, Y., & Wang, E. (2014). R package:
Forecast. Retrieved January 30, 2018, from http://cran.r-
project.org/web/packages/forecast/index.html
Imhoff, M., Lawrence, W., Stutzer, D., & Elvidge, C. (1997).
A technique for using composite DMSP-OLS “City
Lights” satellite data to map urban area. Remote
Sensing of Environment, 61, 361–370. doi:10.1016/
S0034-4257(97)00046-1
Jing, X., Shao, X., Cao, C., Fu, X., & Yan, L. (2016).
Comparison between the Suomi-NPP day-night band
and DMSP-OLS for correlating socio-economic variables
at the provincial level in China. Remote Sensing, 8, 17.
2016 doi:10.3390/rs8010017
Kalogirou, S. (2003). The statistical analysis and modelling of
interval migration flows within England and Wales (Doctoral
dissertation). Newcastle University Library (201 29808 1).
Retrieved January 30, 2018, from gisc.gr/?mdocs-file=1245
Letu, H., Hara, M., Tana, G., & Nishio, F. (2010). A saturated
light correction method for DMSP/OLS nighttime satellite
imagery. IEEE Transactions on Geoscience and Remote
Sensing, 50, 389–396. doi:10.1109/TGRS.2011.2178031
Li, X., Huimin, X., Xiaoling, C., & Chang, L. (2013).
Potential of NPP-VIIRS nighttime light imagery for
modeling the regional economy of China. Remote
Sensing, 2013 5 3057–3081. doi:10.3390/rs5063057
Liang, S., Yu, Y., & Defelice, T.P. (2005). VIIRS narrowband
to broadband land surface albedo conversion: Formula and
validation. International Journal of Remote Sensing, 26(5),
1019–1025. doi:10.1080/01431160512331340156
Lu, D., Tian, H., Zhou, G., & Ge, H. (2008). Regional mapping
of human settlements in southeastern China with multisen-
sor remotely sensed data. Remote Sensing of Environment,
112, 3668–3679. doi:10.1016/j.rse.2008.05.009
Milaka, K.G. (2010). Spatial analysis of urban area’s devel-
opment through multivariable alternative scenario proces-
sing (Doctoral dissertation). Department of Planning
and Regional Development, University of Thessaly.
Retrieved January 30, 2018, from http://thesis.ekt.gr/
thesisBookReader/id/19722#page/14/mode/2up
EUROPEAN JOURNAL OF REMOTE SENSING 245
Miller, S.D., Straka, W., Mills, S.P., Elvidge, C.D., Lee, T.F.,
Solbrig, J., … Weiss, S.C. (2013). Illuminating the cap-
abilities of the suomi national polar-orbiting partnership
(NPP) visible infrared imaging radiometer suite (VIIRS)
day/night band. Remote Sensing, 5(12), 6717–6766.
doi:10.3390/rs5126717
Mukim, M., & Garrett, K. (2013). Tracking light from
space: Innovative ways to measure economic develop-
ment. Retrieved January 30, 2018, from http://blogs.
worldbank.org/sustainablecities/tracking-light-space-
innovative-ways-measure-economic-development
Murak, J. (2013). Regression analysis using GIS. Retrieved
January 30, 2018, from https://libraries.mit.edu/files/gis/
regression_presentation_iap2013.pdf
National Oceanic and Atmospheric Administration-NOAA
(DMSP). (2018). Retrieved January 30, 2018, from
https://ngdc.noaa.gov/eog/dmsp/
downloadV4composites.html
National Oceanic and Atmospheric Administration-NOAA
(VIIRS). (2018). Retrieved January 30, 2018, from https://
ngdc.noaa.gov/eog/viirs/download_dnb_composites.html
Organisation for Economic Co-operation and Development
(OECD). (2018). OECD tourism trends and policies 2018.
Retrieved January 18, 2019, from https://www.oecd-ili
b r a r y . o r g / d oc s e r v e r /t ou r - 2 0 1 8 e n . p d f ? e x p i r e s =
1548600428&id=id&accname=guest&checksum=
43193AFD331298E4D63B00B5E4BF61EE
Papatheodorou, A., & Arvanitis, P. (2014). Tourism and
the economic crisis in Greece - regional perspectives.
Retrieved January 18, 2019, from http://region-develop
pement.univ-tln.fr/en/pdf/R39/9-Papatheodorou.pdf
Qingling, Z., Levin, N., Chalkias, C., & Letu, H. (2015).
Nighttime light remote sensing: monitoring human
societies from outer space. In P.S. Thenkabail (Ed.),
Remote sensing handbook -volume III: Water resources,
disasters, and urban: Monitoring, modeling, and mapping
(pp. 289–311). CRC Press. Taylor and Francis Group.
ISBN: 139781482217926.
Rayner, J., Raupach, M.R., Paget, M., Peylin, P., & Koffi, E.
(2010). A new global gridded data set of CO2 emissions
from fossil fuel combustion: Methodology and evalua-
tion. Journal of Geophysical Research: Atmospheres
(1984–2012), 115(D19). doi:10.1029/2009JD013439
Shi, K., Huang, C., Yu, B., Yin, B., Huang, Y., & Wu, J.
(2014a). Evaluation of NPP-VIIRSnight-time light com-
posite data for extracting built-up urban areas. Remote
Sensing Letters, 5, 358–366.
Shi, K., Yu, B., Huang, Y., Hu, Y., Yin, B., Chen, Z., … Wu,
J. (2014b). Evaluating the ability of NPP-VIIRS night-
time light data to estimate the gross domestic product
and the electric power consumption of China at multiple
scales: A comparison with DMSP-OLS data. Remote
Sensing, 6, 1705–1724. doi:10.3390/rs6021705
Small, C., Pozzi, F., & Elvidge, C.D. (2005). Spatial analysis
of global urban extent from DMSP-OLS night lights.
Remote Sensing of Environment, 96, 277–291.
doi:10.1016/j.rse.2005.02.002
Stathakis, D., & Baltas, P. (2017). Seasonal population
estimates based on night-time lights. Computers,
Environment and Urban Systems, 68, 133–141. March
2018 doi:10.1016/j.compenvurbsys.2017.12.001
Stevenson, N., Aiery, D., & Miller, G. (2008). Tourism policy
making: The policymakers’ perspectives. Retrieved January
30, 2018, from http://epubs.surrey.ac.uk/1118/1/fulltext.pdf
Sutton, P. (1997). Modelling population density with night-
time satellite imagery and GIS. Computers, Environment
246 E. KRIKIGIANNI ET AL.
and Urban Systems, 21(3–4), 227–244. May–July 1997
doi:10.1016/S0198-9715(97)01005-3
Sutton, P., Roberts, D., Elvidge, C., & Baugh, K. (2001).
Census from heaven: An estimate of the global human
population using night-time satellite imagery.
International Journal of Remote Sensing, 22(16), 3061–
3076. doi:10.1080/01431160010007015
UNWTO. (2017). Tourism highlights: 2017 Edition. ISBN:
978-92-844-1901-2, Retrieved January 30, 2018, from
https://www.e-unwto.org/doi/book/10.18111/
9789284419029
Wei, Y., Hongxing, L., Wei, S., Bailang, Y., & Chunliang,
X. (2014). Normalization of time series DMSP-OLS
nighttime light images for urban growth analysis with
Pseudo invariant features. Landscape and Urban
Planning, 128, 1–13. August 2014 doi:10.1016/j.landur
bplan.2014.04.015
Zhao, N., Hsu, F., Cao, G., & Samson, E. (2017).
Improving accuracy of economic estimations with
VIIRS DNB image products. International Journal
of Remote Sensing. doi:10.1080/01431161.2017.133
1060